frames

app.py

@@ -32,10 +32,7 @@ try:
 
     sh(f"pip install {flash_attention_wheel}")
     print("Attempting to download and install FlashAttention wheel...")
-    # sh("pip install flash-attn")
-    sh("pip install --no-build-isolation transformer_engine-2.5.0+f05f12c9-cp310-cp310-linux_x86_64.whl")
 
-    # tell Python to re-scan site-packages now that the egg-link exists
     import importlib, site; site.addsitedir(site.getsitepackages()[0]); importlib.invalidate_caches()
 
     flash_attention_installed = True
@@ -54,7 +51,6 @@ try:
     sh(f"pip install {te_wheel}")
     print("Attempting to download and install Transformer Engine wheel...")
 
-    # tell Python to re-scan site-packages now that the egg-link exists
     import importlib, site; site.addsitedir(site.getsitepackages()[0]); importlib.invalidate_caches()
 
 except Exception as e:
@@ -123,10 +119,10 @@ def calculate_required_time(steps, max_duration):
     warmup_s = 60
 
     max_duration_duration_mapping = {
-        1: 8,
-        2: 8,
-        3: 12,
-        4: 20,
+        20: 8,
+        45: 8,
+        70: 12,
+        95: 20,
     }
     each_step_s = max_duration_duration_mapping[max_duration]
     duration_s = (each_step_s * steps) + warmup_s
@@ -247,12 +243,6 @@ def run_pipeline(prompt_text, steps, image_paths, audio_file_path, max_duration
     filename = f"gen_{uuid.uuid4().hex[:10]}"
     width, height = 832, 480
 
-    duration_frame_mapping = {
-        1:25,
-        2:45,
-        3:70, 
-        4:97
-    }
 
     # Run inference
     runner.inference_loop(
@@ -265,7 +255,7 @@ def run_pipeline(prompt_text, steps, image_paths, audio_file_path, max_duration
         width,
         height,
         steps,
-        frames = int(duration_frame_mapping[max_duration]),
+        frames = int(max_duration),
         tea_cache_l1_thresh = 0.0,
     )
 
@@ -345,7 +335,7 @@ with gr.Blocks(css=css) as demo:
         default_steps = 10
         default_max_duration = 3
 
-        max_duration = gr.Slider(minimum=2, maximum=4, value=default_max_duration, step=1, label="Max Duration")
+        max_duration = gr.Slider(minimum=45, maximum=95, value=default_max_duration, step=25, label="Frames")
         steps_input = gr.Slider(minimum=10, maximum=50, value=default_steps, step=5, label="Diffusion Steps")
         
 
@@ -376,7 +366,7 @@ with gr.Blocks(css=css) as demo:
         gr.Markdown("")
         time_required = gr.Markdown(get_required_time_string(default_steps, default_max_duration))
         run_btn = gr.Button("🎬 Action", variant="primary")
-        
+
         gr.Examples(
             examples=[
 
@@ -386,7 +376,7 @@ with gr.Blocks(css=css) as demo:
                     10,
                     ["./examples/naomi.png"], 
                     "./examples/science.wav", 
-                    3,              
+                    70,              
                 ],
 
                 [
@@ -394,15 +384,15 @@ with gr.Blocks(css=css) as demo:
                     10,
                     ["./examples/art.png"], 
                     "./examples/art.wav", 
-                    2,              
+                    45,              
                 ],
 
                 [
                     "A handheld tracking shot follows a female warrior walking through a cave. Her determined eyes are locked straight ahead as she grips a blazing torch tightly in her hand. She speaks with intensity.",
-                    5,
+                    10,
                     ["./examples/naomi.png"], 
                     "./examples/dream.mp3",     
-                    5,             
+                    95,             
                 ],
 
                 [
@@ -410,7 +400,7 @@ with gr.Blocks(css=css) as demo:
                     40,
                     ["./examples/amber.png", "./examples/jacket.png"],
                     "./examples/fictional.wav",     
-                    4,             
+                    70,             
                 ],
 
             ],